On Wed, Jan 02, 2019 at 11:37:56AM -0800, Andrei Vagin wrote:
> + if (unlikely(ret)) {
> + dput(fc->root);
> + deactivate_locked_super(fc->root->d_sb);
> + fc->root = NULL;
Er... I'd rather not dereference the damn thing after dput().
It's _probably_ OK (fc->root here, AFAICS, is going to be
equal to its ->sb->s_root, and thus still pinned), but it's
better to fetch ->d_sb first. If nothing else, it's easier
to prove correctness that way...
